implementing self-organizing information … · and self-synchronization of autonomous and...

International Journal of InnovativeComputing, Information and Control ICIC International c©2014 ISSN 1349-4198Volume 10, Number 2, April 2014 pp. 437–458

IMPLEMENTING SELF-ORGANIZING INFORMATIONDISSEMINATION INTO NETWORK-CENTRIC

MACHINE-TO-MACHINE SYSTEMS

Damjan Katusic and Gordan Jezic

Faculty of Electrical Engineering and ComputingUniversity of Zagreb

Unska 3, 10000 Zagreb, Croatia{damjan.katusic; gordan.jezic }@fer.hr

Received December 2012; revised April 2013

Abstract. This paper explores the convergence of Machine-to-Machine (M2M) systemswith Network-Centric Operations (NCO), and analyzes the perspective of integrating self-organizing gossip protocols as an information dissemination solution. An overview ofthe M2M communication paradigm and current standardization efforts regarding archi-tecture are given. The paper includes an overview of the network-centric networkingand analyses the idea of implementing NCO approach to achieve situational awarenessand self-synchronization of autonomous and intelligent M2M devices in networked M2Menvironments. Two case studies involving M2M systems are given and discussed as aproof of concept for the implementation of a network-centric approach. Properties ofself-organizing systems are briefly analyzed, and gossip protocols are identified as a vi-able support solution for the proposed convergence.Keywords: Machine-to-machine, Network-centric operations, Self-organization, Infor-mation dissemination, Gossip protocols

1. Introduction. Machine-to-Machine (M2M) systems’ rapid growth and developmentin recent years have attracted much attention. Numerous companies and network op-erators have established specialized M2M departments extending their businesses andproviding new types of automated services. Ubiquitous wireless and wired connectivity,and declining prices of communication modules are the main drivers of such a trend. Thepotential for further growth of the M2M market is enormous: European Telecommunica-tions Standards Institute (ETSI) suggests that the number of connectable machines is fivetimes greater than the amount of humans, although the number of currently connectedmachines is still significantly lower. However, even today the number of connected M2Mdevices is measured in hundreds of millions. Harbor Research expects 390 million cellu-lar M2M connections in 2014 [1], while Mobile Market Development predicts a projectedcompound annual growth rate of 25% through 2014, and approximately 50 billion M2Mdevices worldwide by 2025 [2]. Ericsson’s predictions reach 50 billion M2M devices evensooner, by 2020 [3]. According to Juniper Research, M2M market as a whole, includingboth fixed and mobile technologies, will by 2014 reach value of nearly $35 billion [4].

Both leading M2M standardization bodies (ETSI and Third-generation PartnershipProject (3GPP)) publish similar definition of the M2M communication: it is the commu-nication between two or more entities with little or no direct human intervention [5,6].Similar definitions can be found in numerous other publications, books or articles dealingwith the same matter. From the functional point of view, a communication system basedon M2M interactions typically comprises of geographically dislocated devices, communi-cation channels that connect them and a platform for management functions. Actors in

437

438 D. KATUSIC AND G. JEZIC

such an environment can include broad range of communication capable devices: comput-ers, mobile phones, tablets, but also variety of sensors, smart grid networks, embeddedprocessors, industrial and medical equipment, and countless other devices. M2M technol-ogy offers various applications and a vast potential for future development of new typesof applications. One of the unofficial divisions proposed in [7] identifies six applicationareas: building management, transportation and logistics, healthcare, local communitiesand public safety, energy, manufacturing, and industrial applications.As stated in [8], M2M technology is based on the idea that a machine has more value

when it is networked and that the network becomes more valuable as more machines areconnected. Operation and control in large and dynamic M2M systems, as is the case withdistributed systems in general, is still an emerging area of research. This paper extendsthe ideas suggested in [9] where the convergence of network-centric networking approachinto machine-to-machine environments was discussed. Network-centric operation and con-trol offers a scalable decentralized alternative to centralized control and its typical issues(scalability, traffic congestion (“bottleneck”), transmission delays, energy wastage, etc.)[10]. Enabling qualitative information management between communicating machines isthe necessary prerequisite for information sharing (“bringing the right information to theright destination at the right time”), stimulating machine cooperation, and generally im-proving individual and collective machine operations. Effective information disseminationenables faster machine operations and ultimately better performance of the M2M systemas a whole. Self-organization principles based on the natural systems provide an attrac-tive approach for handling requirements in dynamic distributed environments, especiallyfor those M2M systems that emphasize machine autonomy and cooperation.This paper is organized as follows. In Section 2, we give an overview of the current stan-

dardization efforts in the M2M domain, while Section 3 brings prominent considerations inthe area of M2M architectures, with the prospect of implementing device-to-device com-munication in standardized architecture layouts. Section 4 focuses on the integration ofnetwork-centric approach into an M2M system and discusses its potential importance forM2M communication, particularly information sharing between M2M devices. In Section5, we give an analysis of two M2M scenarios in the context of analyzed network-centricnetworking. Section 6 brings an overview and a brief analysis of the bio-inspired self-organization principles that can be applied in a network-centric M2M environment, withparticular focus on the gossip algorithms developed for scalable information disseminationand aggregation. Finally, Section 7 concludes the paper.

2. M2M Standardization Efforts. Current networks are optimized for Human-to-Human (H2H) interactions and data transfer, and communication patterns in such systemscan greatly differ from those in M2M systems. It is important that communication tech-nologies evolve and develop capabilities to efficiently support both human and machinesolutions without impairing their capabilities.An M2M concept is not a revolutionary idea since it has been present in various forms

over the years. Early 1990s witnessed the development of Supervisory Control and DataAcquisition (SCADA) systems [11]. These primitive industrial management and telemetrysystems are usually perceived as precursors of modern M2M systems. They include manyconnected sensors from which they gather data. However, their biggest disadvantage ishigh cost because they are based on proprietary communication technologies. UnlikeSCADA, M2M systems work with standardized technologies [8] and are independent ofthe used access in a wide range of possible wired/wireless network solutions (DigitalSubscriber Line (DSL), Wi-Fi, ZigBee, Bluetooth, cellular, satellite, etc.).

IMPLEMENTING SELF-ORGANIZING INFORMATION DISSEMINATION 439

Many standardization bodies, including mentioned ETSI and 3GPP, have recently en-gaged in M2M standards development: International Telecommunication Union – Telecommunication Standardization Sector (ITU-T), Institute of Electrical and Electronics En-gineers (IEEE), 3rd Generation Partnership Project 2 (3GPP2) and TelecommunicationsIndustry Association (TIA) [12]. Considering the positive impulse of M2M standardiza-tion activities in recent years and learning on the experiences of 3GPP, ETSI is currentlyjoining six other major standardization organizations from around the world (Associationof Radio Industries and Businesses (ARIB) in Japan, Alliance for TelecommunicationsIndustry Solutions (ATIS) and TIA in the USA, China Communications Standards As-sociation (CCSA) in China, and Telecommunications Technology Association (TTA) inSouth Korea) to form a global M2M partnership: oneM2M [13]. Its agenda is to encouragethe development of a common M2M Service Layer that can be embedded within varioushardware and software, and relied upon to connect a variety of devices in the field withM2M application servers worldwide.

ETSI produces globally applicable standards for Information and CommunicationsTechnology (ICT) including fixed, mobile, and radio communications, Internet, and otherareas, and continually strives to improve collaboration with other research bodies [14]. In2007, a new ETSI Technical Committee (TC) for developing standards for M2M commu-nication (Table 1) has been established [15]. The group aims to provide an end-to-endview of M2M standardization. Their work regarding M2M technology has begun withstandards that analyze different use cases: smart metering [16], eHealth [17], connectedconsumer [18], automotive applications [19], and city automation [20]. The objective is tocover enough prevailing use cases to ensure that all of the important requirements of M2Msystems are identified so that the associated architecture work provides the foundationfor future M2M applications. Anyhow, they have also put some effort into defining basicM2M terminology [21], as well as its service requirements [5] and functional architecture[22]. ETSI’s cooperation with other organizations has also resulted in few published stan-dards. They closely co-operate with the work of the 3GPP [23] and 3GPP2 [24] standardsinitiatives for mobile communication, then with Open Mobile Alliance (OMA) [25] andBroadband Forum (BBF) [26] regarding their solutions for remote management, and withthe ZigBee Alliance on the subject of interworking with their area networks [27].

3GPP produces highly successful reports and specifications that define 3GPP tech-nologies, from Global System for Mobile Communications (GSM), General Packet RadioService (GPRS), and Enhanced Data rates for GSM Evolution (EDGE) towards UniversalMobile Telecommunications System (UMTS), and recently Long Term Evolution (LTE)[28]. Most of its M2M (to avoid possible confusion it is important to mention that 3GPPuses term Machine-Type Communication or abbreviated MTC) standardization efforts(Table 1) are conducted within several Service and System Aspects Working Groups (SAWG): SA WG11, SA WG22 and SA WG33. WG1 focuses on services [6,29-32], WG2on architecture [33], and WG3 on problems regarding security [34,35]. Two additionalgroups are actively working on the improvements for radio access networks: GSM EDGERadio Access Network (GERAN) WG24 on protocol aspects of GPRS/EDGE networks[36], and Radio Access Network (RAN) WG25 on Layer 2 and Layer 3 Radio Resourcespecification [37].

1TSG SA WG1 Services, http://www.3gpp.org/SA1-Services2TSG SA WG2 Architecture, http://www.3gpp.org/SA2-Architecture3TSG SA WG3 Security, http://www.3gpp.org/SA3-Security4TSG GERAN WG2, http://www.3gpp.org/GERAN-2-Protocol-Aspects5TSG RAN WG2, http://www.3gpp.org/RAN2-Radio-layer-2-and-Radio-layer


Table 1. ETSI and 3GPP M2M standards

StandardizationDescription

Specificationbody reference

ETSI

M2M Service RequirementsM2M Functional ArchitectureSmart Metering Use CasesM2M DefinitionseHealth Use CasesConnected Customer Use CasesCity Application Use CasesAutomotive Applications Use CasesM2M Interfaces mIa, mId, dIaImpact of Smart Grids on M2M PlatformInterworking with 3GPP NetworksInterworking with 3GPP2 NetworksInterworking with M2M Area NetworksOMA DM Compatible ObjectsBBF TR-069 Compatible Data Model

TS 102 689TS 102 690TR 102 691TR 102 725TR 102 732TR 102 857TR 102 897TR 102 898TS 102 921TR 102 935TR 101 603TR 103 107TR 102 966TS 103 092TS 103 903

3GPP

SA1 – M2M Study ReportSA1 – MTC Service RequirementsSA1 – Alternatives to E.164 for MTCSA1 – Man-machine Interface of the Mobile StationSA1 – Man-machine Interface of the User EquipmentSA2 – System Improvements for MTCSA3 – M2M Security Aspect for Remote Provisioningand Subscription ChangeSA3 – Security Aspect of M2M3GPP Study on RAN Improvements for MTC3GPP Study on GERAN Improvements for MTC

TR 22.868TS 22.368TR 22.988TS 02.30TS 22.030TR 23.888TR 33.812

TR 33.868TR 37.868TR 43.868

3. M2M System Architecture Considerations. Broad market potential of M2Msystems is consequence of their numerous possible applications and use cases, as wellas the variety of available access technologies that can be used in their implementation.These systems need to be reliable, scalable, secure, and manageable. The easiest wayto accomplish this, and to solve set of unique challenges that differentiate them fromH2H systems (large number of connected devices, specific traffic patterns, many types ofdevices, small energy requirements, etc.) is standardization. One of its main aspects isconsideration regarding architecture of M2M systems.

3.1. ETSI architecture standardization approach. ETSI TC M2M published in [22]high-level architecture concept for M2M application support (Figure 1). It includes net-work (Access Network, Core Network, M2M Service Capabilities, M2M Applications, Net-work Management, and M2M Management Functions), and device and gateway domains(M2M Devices, M2M Gateway, and M2M Area Network, including M2M Applicationsand M2M Service Capabilities). M2M Devices include broad range of devices cited in theintroduction, run M2M Application(s) using M2M Service Capabilities, and are capableto autonomously (without human intervention) exchange data with other devices. Theyconnect to network domain in two different ways: directly or through M2M Gatewaywhich serves as a network proxy. M2M Gateway using M2M Service Capabilities to en-sure M2M Devices are interworking and interconnected to the underlying communication


Figure 1. ETSI M2M system architecture

network, and can provide various services to them. M2M Area Network connects gate-way and devices that lack service capabilities and are not capable to directly connectto Access Network. Network Domain comprises of Access and Core Networks, enablescommunication between M2M Gateways and M2M Applications, and includes Networkand M2M Management Functions. Access Networks include (but are not limited to):DSL, satellite, GERAN, Universal Terrestrial Radio Access Network (UTRAN), evolvedUTRAN (eUTRAN), Wireless Local Area Network (WLAN), and Worldwide Interoper-ability for Microwave Access (WiMAX). Core Networks (CN) include (but are not limitedto): 3GPP CNs, ETSI Telecoms and Internet converged Services and Protocols for Ad-vanced Networks (TISPAN) CN, and 3GPP2 CN. M2M Applications run the service logicand use M2M Service Capabilities accessible via an open interface.

There are three reference points between various node pairs (M2M Device, M2M Gate-way, and Network (e.g., Application Server)) that define ETSI M2M functional architec-ture framework (Figure 2):

• The mIa reference point offers generic and extendable mechanism for Network Ap-plication (NA) interactions with the Network Service Capabilities Layer (NSCL).

• The dIa reference point offers generic and extendable mechanism for Device Appli-cation (DA)/Gateway Application (GA) interactions with the DSCL/GSCL. Whenbetween DA and GSC, typically over an Internet Protocol (IP) network, it is im-plemented over Constrained Application Protocol (CoAP) or Hypertext TransferProtocol (HTTP).

• The mId reference point, between SCLs (DSCL/GSCL to NSCL), offers generic andextendable mechanism for SCL interactions. When between GSC and NSC, it isimplemented over HTTP or Session Initiation Protocol (SIP).

It can be observed from the figure that M2M Device connects to Network Domaineither directly (includes Service Capability Layer) or through the M2M Gateway (lacksSCL). ETSI so far in its standards does not assume direct device-to-device communicationscenario (apart from the local interactions inside the M2M Area Network), but everycommunication involves Network Domain and associated servers who are then responsiblefor providing connectivity and/or other services to M2M Gateway(s)/Device(s).


Figure 2. M2M functional architecture framework

Figure 3. 3GPP M2M system architecture

3.2. 3GPP architecture standardization approach. Recent developments in wire-less industry such as widespread availability of wireless connectivity, declining prices ofM2M modules, and regulatory incentives for certain industries (smart grid, e-Health)have attracted attention of potential M2M stakeholders [12]. With the merit of providinghigher-layer connections, 3GPP network scenarios have been regarded as one of the mostpromising M2M solutions [11].MTC Device comprises of variety of devices mentioned in the introduction and connects

via MTCu interface to the 3GPP network (GERAN, UTRAN, eUTRAN, etc.). Dependingon the used access technology, MTCu can be based on one of the following interfaces: Uu,Um, Ww or LTE-Uu. The 3GPP network provides transport and communication services(including 3GPP bearer services, IP Multimedia Subsystem (IMS), and Short MessageService (SMS)) optimized for the MTC communication that connect MTC Device withan MTC Server or other MTC Devices. The MTC Server connects to the 3GPP networkvia MTCi/MTCsms interface and thus communicates with MTC Devices [29] (Figure 3).There are three scenarios regarding the communication between MTC Servers and MTC

Devices [6]. MTC Device indicates wireless MTC modules and terminals connected via anaccess network, included in the “machines in the field” domain. MTC Server stands forcentral servers that communicate with MTC Devices, and can be located inside or outsideMobile Network Operator (MNO) domain, or within public Internet. First communicationscenario involves MTC Devices, distinguishable from each other, communicating with one


MTC Server (N to 1), while in second scenario there are several MTC Servers and severalMTC Devices (N to N). In first scenario, one MTC module communicates with one serveronly, while second scenario involves more servers for load distribution. Third scenarioconsists of MTC Devices communicating directly with each other without the intermediateMTC Server. The latter scenario is not seen as the relevant within the scope of 3GPP’swork on M2M, because most popular use cases involve some kinds of M2M device-to-servercommunication. Therefore, 3GPP and ETSI seem to have a rather similar standpointregarding Machine-to-Machine communication between devices without intermediaries,and leave this area open to further discussion and research. However, there is no doubtthat certain methods and principles, such as the ones discussed in the following chapters(e.g., network-centric approach), may greatly improve specific M2M applications.

4. Network-Centric Approach in M2M Systems. M2M systems are due to theirdiversity still faced with information management and coordination challenges. They canconsist of various types and numbers of devices, local or wide area coverage, differentlevel of mobility, autonomy, intelligence or energy constraints. Information managementin such an environment that can be based on one of the two basic types, centralized ordecentralized. Centralized information management system is continuously examining theenvironment using its sensor capabilities, transmitting the gathered data to the centralnode (M2M/MTC Server) for processing, and eventually decision making. Commands aredistributed from centre to the edges [10]. This approach is correlated to the M2M/MTCDevice(s)-M2M/MTC Server(s) 3GPP scenarios. In decentralized information manage-ment approach, inherent in the NCO principles, control and intelligence is shifted fromone node to the whole network [10], or as it is in [38] called, “infostructure”. The decen-tralization of control and intelligence among many nodes allows processing of gathereddata within the network, which is in relation to the M2M/MTC Device-M2M/MTC De-vice 3GPP scenarios. NCO features motivate the idea of integrating some of its conceptsin the perspective area of machine type communications.

4.1. Network-centric operations. Network-Centric Operations (NCO) refer to a con-tinuously evolving, complex community of people, devices, information, and services in-terconnected by a communications network in order to optimize resource managementand provide superior information on events and conditions needed to empower decisionmakers [39]. The term “network-centric” is attributed to the U.S. Admiral Jay Johnsonwho used it to describe what he has seen as a “fundamental shift from platform-centricwarfare to network-centric warfare” [38]. Network-centric operations approach is in directopposition to platform centricity, it shifts from viewing actors as independent to view-ing them as part of a continuously adapting ecosystem, and emphasizes the importanceof making strategic decisions to adapt or even survive in such a changing environment[38,40]. Its underlying framework has been influenced by certain command and controlprocesses characterized by an iterative sequential series of steps, such as Observe, Ori-ent, Decide and Act (OODA) loop cycle attributed to former United States Air Force(USAF) Colonel John Boyd, a model consisting of sense, process, compare, decide, andact steps, developed by Dr. J. S. Lawson, and the Headquarters Effectiveness AssessmentTool (HEAT) process developed by Dr. R. E. Hayes in 1984 [40].

Network-centric (sometimes also referred to as information-centric or knowledge-centric)approach seeks to achieve an information advantage, enabled in part by information andcommunication technologies, into a competitive advantage through the robust networking.Specifically, the NCO concept contains the following four tenets, as proposed in [41]:

• A robustly networked force improves information sharing.


• Information sharing enhances the quality of information and shared situational awar-eness.

• Shared situational awareness enables collaboration and self-synchronization, and en-hances sustainability and speed of command.

• These, in turn, dramatically increase work (mission) effectiveness.

These tenets can be, with certain adjustments, applied to any networked environment.For a success, especially in a dynamic environment, it is critical to achieve informationsharing that enables shared situational awareness and self-synchronization. This, simplyspeaking, means to deploy as much information about relevant aspects of the networkedenvironment (e.g., the position and availability of critical resources, status of neighboringnodes, position of link/node failures, traffic congestions.) to nodes that need to know itin order to achieve better work performance in a dynamic distributed environment.

4.2. Network-centric operations in M2M systems. NCO paradigm described in theprevious section mentioned several important features. However, despite the necessarytechnological basis that allows all of these features to fit together, the most importantvariable is behavior of a decision maker and its ability to make use of acquired informationand knowledge. People, characterized by their cognitive processes and abilities, as wellas their social interactions and organization, are the final destination of almost all of theacquired information and knowledge through networking. NCO results are ultimatelydriven and evaluated by human reasoning. Nodes in a network-centric environment areautonomous decision making units that can collaborate with other nodes (serve otherunits or be served by them). Some machines in M2M environment are capable of sametype of behavior: function autonomously, without the direct human intervention, andare characterized by different cognitive and learning abilities. In such an environmentthey do not only support the network to process data for human decision makers, butare at some level decision makers themselves. They are driven to adapt to changes intheir environments and reach goals through autonomous social interactions with othermachines and autonomous decision making.Information sharing. Information exchange is an important aspect of any networked

environment. Nodes (i.e., people, connected machines) collect information through in-teractions with other nodes and use it to solve operational tasks and enrich their ownknowledge. M2M systems with potentially large number of nodes offer large quantities ofinformation that can be used for their own benefit. Establishing collaboration betweenconnected machines assumes a bit more than interoperability, which is a fundamentaltechnical ability to communicate. It is defined on a more abstract level than protocolsthat allow communication and adds semantic layer to the data that is being exchanged,encouraging “meaningful” conversations to take place. Semantic description of both dataand networked M2M machines that exchange it is the first step to establishing machineautonomy and collaboration, although human level of thinking and understanding ofexchanged knowledge will not be achieved for years to come. Standardization in M2Msystems has started in this area as well [42], and it will be interesting to see how connectedM2M machines, despite the fact that many of them are constrained by their processorand memory capabilities, will be able to “talk” and understand each other.Further insight into the importance of information sharing is provided by Metcalfe’s

Law, which describes the potential value of a network. It states that as the number ofnodes in a network (n) increases linearly, the potential “value” or “effectiveness” (v) ofthe network increases nonlinearly as the square number of nodes [40] (Figure 4). Thesource of potential network value is a function of the interactions between the nodes. Forevery n = N nodes in a network, there are N − 1 potential interactions between them.


Figure 4. Metcalfe’s law

Therefore, the potential value of the network is: v = N × (N − 1) = N2 −N . For largeN , the potential value scales with v = N2.

Therefore, more nodes mean more information that can be used to achieve set goals.The NCO paradigm was devised in the first place to allow human users to extract qualita-tive information from the networked environment for better decision making. Situation isthe same if M2M devices are the decision makers: network-centricity implies the necessityfor better information connectivity between M2M network nodes in such an environment.There are numerous approaches for implementation of a decentralized information dissem-ination system, and some suggestions are going to be proposed in the following chaptersof this paper.

Self-synchronization. Self-synchronization is the notion that highly complex groups,with accurate detailed information available at all levels, organize naturally (and opti-mally) from the bottom-up [43]. The term self-synchronization proposed in the network-centric literature is closely related to the more general term of self-organization whichdescribes the class of processes that occur in a variety of different systems, and are allcharacterized by the bottom-up arising order through local interactions between the com-ponents of the system. Such an organizational principle is inherent in many aspects ofhuman societies or numerous animal groups (e.g., ant colonies, swarms, flocks of birds),and can also be achieved in M2M systems. Then it creates new operational capabilities(machines are able to make decisions based on information gathered through sensors, fromother machines, from databases, etc.) through autonomous cooperation of machines andallows better decision making.

Decision making. Finally, the last important aspect of NCO, and the desired con-sequence of all mentioned features, is the smart decision making. It implies that a nodein a networked environment has access to underlying information and that it uses it forconducting both regular or out of the ordinary work activities. Human societies are ac-customed to hierarchical organization and highly-centralized top-down commands. NCOchanges that notion and pushes shared situational awareness to the edge of the network.This disrupts traditional practices in top-down command and control environments, es-pecially military [44], which is in fact one of the pioneers of NCO research. Firstly,information non-attribution reverses the assumption that commands are issued from anindividual entity to a particular individual entity: they are issued to a pool with ratherundefined responsibilities. Secondly, as a consequence of such decentralization, decisionmaking is migrating to the edges of a network, giving access to information of qualityand quantity that is potentially equal to or better than that available at the centre. US


Army’s restructuring into smaller units, such as formation of Stryker brigades [45], is onevidence that even strict hierarchical human organizations have started to accept benefitsof such an approach. Decentralization can also be applied in M2M environment, nev-ertheless ETSI and 3GPP either so far prefer architectures with servers who act as thecentral points of control over the group of connected machines. Such an approach willallow machines to enrich their capabilities and achieve full utilization of the informationavailable in the system. One of the main prerequisites for accomplishing this is an imple-mentation of a robust information grid within the M2M system that fosters cooperationand allows sharing of structured data, information, and knowledge.

5. Case Study: Network-Centric M2M Scenarios. This chapter brings forward theanalysis of integration of network-centric ideas into two possible M2M scenarios.Healthcare M2M systems will be according to projections in [1] one of the main market

drivers of the global M2M growth in the following years. Healthcare is an information richand knowledge intensive environment, and in order to treat and diagnose even a simplecondition, a physician must combine many varied data elements and information [46].The analysis of the first scenario presumes that the architectural framework of the

observed M2M e-Health scenario is largely based on the ETSI Remote Patient Monitoringmodel drafted in [17], with the inclusion of 3GPP cellular network as its underlyingnetwork access and core solution. Such a system includes remote patient monitoringM2M Devices connected via a Home Hub which acts as an M2M Gateway to the clinicalside. Patient device gathers patient measurements, which may be communicated eachtime device gathers data, accumulated measurements may be communicated periodically(e.g., hourly or daily), or data may be delivered upon request or upon certain events.Clinical side involves care coordinator M2M services that monitor received patient data(M2M Server), and if measurements indicate that there has been a change in the patient’shealth status, or fall outside of a predetermined range, they alert clinician personnel(physician, nurse, etc.). This study for the more complete e-Health picture also involvessome other healthcare stakeholders that are not crucial for the remote monitoring case,such as pharmacy and healthcare insurance (Figure 5).Presently, many existing M2M e-Health initiatives represent distinctive and loosely

connected entities whose operation is still largely platform-centric, that is, concentrateson the operations of a single subsystem (platform) with too little regard for the operational

Figure 5. M2M-NCO e-Health scenario


interaction among different subsystems. The basic service of remote patient monitoringwith remote M2M Devices and a central M2M Server is set up properly, so there is noneed to change such an architectural decision. However, the inclusion of network-centricoperations could improve analyzed scenario, stimulate better cooperation between variousactors in such an environment (e.g., M2M remote monitoring devices, M2M Gateways,databases that maintain local measurement records), and include new types of services(e.g., automatically order new batch of medicines from pharmacy when current reservesbecome exhausted) without the need for an inclusion of a server side.

Second analyzed scenario involves a rather plausible system of M2M Devices equippedwith various sensors scattered on a farming field. They could be used for measuring variousfarming relevant parameters such as soil temperature, moisture, chemical composition ofthe soil, and reporting it to the M2M coordinator service (M2M Server). One of themain prerequisites in such a system is the simplicity of its M2M end-devices, especiallyin the context of energy efficiency. Obviously, devices that would be put in soil shouldnot be too big in dimensions, and they would need to be able to operate for a reasonableamount of time. Switching batteries is almost not applicable in this situation, and sensorcapabilities that are the basis of a solution already consume an important part of availableenergy. The communication segment should be carefully designed: M2M Devices that arescattered on a possibly large field need to be able to establish a communication to server,gateway and each other. Solutions such as satellite or mobile cellular networks that offerlarge coverage are a possibility, but one has to take into account the energy cost. Also,fields can be located on remote and hardly accessible terrains with poor communicationinfrastructure that could have problems in supporting a very large number of end-devices.Nearby M2M Gateways would like in the previous scenario serve as proxies to the serverside, and could be equipped with such communication technologies because they are notconstrained as the end-devices. Next possibility is usage of Personal Area Network (PAN)ZigBee M2M modules. Such a solution is cheaper in the context of battery consumptionthan the latter proposals, especially when taking into account possibly large number ofconnected M2M end-devices. ZigBee offers a rather small coverage area, but end-devicescan establish a mesh network between them that would support local interactions andeventually a connection to gateway, while gateway would be responsible for establishingthe connection to the M2M Server (Figure 6). Each M2M Device is able to connectdirectly to its neighbors (full line), and indirectly to all other nodes (broken line). Eachend-device serves as a relay for all other end-devices.

Farm field M2M scenario is very similar to the e-Health scenario described above.However, there are several important differences, apart from the fact that they are tied todifferent application domains. Cooperation between end-devices without the intermediaryis because of the nature of used technology a necessity even for simple operations suchas an exchange of data (e.g., between source and destination M2M Devices highlightedin Figure 6). This scenario is based on an ETSI proposed M2M Area Network template,and is still in accordance with its proposed architecture described earlier in the paper.However, it also offers a very good insight into the limitations of current ETSI and 3GPPstandards: if observed sensors are replaced with a bit more complex M2M end-devices theywould not fit in the proposed architectural models, i.e., current standardization effortswould not be able to properly explain and analyze them. It is important to emphasizethat certain M2M services operate in a dramatically different way. Therefore, apart fromthe described M2M e-Health system where a network-centric networking, although not acompletely natural fit, offers enrichment of the current system’s operations, there are alsoM2M systems with inherently decentralized organizations that are so far mostly ignoredby the standardization and would benefit even more from the inclusion of NCO ideas.


Figure 6. M2M farm field scenario

Mentioned platform-centricity has major influence on the access range and sharing ofinformation stored among (or even within) the existing individual subsystems, and callsfor another approach: implementing a network-centric system that facilitates informationsharing among all participants within the operational space. Such an approach proposesthe creation of an information exchange grid that will allow free flow of information, inobserved cases among healthcare actors or sensors on a large farming field. This approachpushes critical information and shared situational awareness to nodes that need it tosuccessfully accomplish tasks, no matter where are they located in the network.Satisfying “right information to the right place at the right time” concept is not a

trivial task. M2M systems are possibly heterogeneous and dynamic environments thatcould greatly benefit from the establishment of a network-centric and collaborative in-formation management system. There are few important capabilities that such systemshould support: universal access to information (from a variety of sources), orchestratedinformation bus (that connects all M2M Devices and possibly aggregates, filters, andprioritizes information delivery), continuous adaptation to changes (to dynamic networktopologies, M2M system membership changes, etc.), and support defined QoS policiesand mechanisms [47].

6. Self-Organization. Self-organization has been a subject of numerous discussions andresearch since the time of ancient Greece. Various philosophy schools, and later branchesof natural or social sciences have tried to introduce a precise definition of self-organization.Intuitively, self-organization refers to the autonomous arrangement of parts of a systemin a non-random way. It is a result of local interactions and internal constraints that arenot influenced or controlled by anything outside the system. As a result of these non-deterministic and local interactions occurs a phenomena known as emergence. Emergentphenomena as an externally (outside of a system) observable outcome can appear in theform of a particular pattern, property or a behavior. Although it can include rathercomplex patterns and behaviors, typically it is a result of simple interactions that occurwithin a system and without any apparent central control. Self-organizing systems are


encountered in many application areas, and because of the interesting properties theydemonstrate, today they are a subject of numerous scientific efforts.

6.1. Properties of self-organization. Self-organizing systems are characterized by anumber of mandatory and optional properties that define them [48]. Three mandatoryproperties are:

• Global organization: Process of self-organization brings system into relatively stablestate in which it performs its functions. The resulting state can be either staticor stationary. Static case presuming resulting organization positions in a system isfixed, while in the latter (stationary) case organizational components continuouslychange their position, but according to some ordered and stable pattern.

• Dynamic adaptation: Self-organizing systems are capable of dynamically reorganiz-ing their organization in order to adapt to changes either in their functions or theconditions they face from the environment.

• Lack of external control: The dynamic reorganization and resulting order mentionedin the last entry is executed endogenously, i.e., without any external control.

Example of the static global organization is a fixed light bulb which serves its purposeof bringing light to the room when needed. Typical example of a stationary system areBenard cells [49] where pattern remains stable as long as the heating and cooling of theliquid are not altered, so an upward flow of the liquid on one side and a downward streamon the other side are in balance. Dynamic adaption is very clearly demonstrated in theexample of insect populations that collect food, and behave differently according to theenvironmental conditions around them. Before mentioning some of the optional prop-erties of self-organization, especially those important in the context of network-centricnetworking approach, it is important to distinguish its two subtypes: strong and weakself-organization. Strong self-organization systems’ decisions are distributed among sys-tem components without any central control loop. Linking this discussion to the previouschapters, such properties can be achieved in systems that are so far outside of the currentstandardization efforts of ETSI and 3GPP, but we feel obliged to warn about their unat-tended potential. Weak self-organized systems, from an internal point of view, achieveself-organization as a consequence of a centralized planning and control. Scenario aboutM2M sensors distributed on a farm field is a very good fit for the latter. Although certainaspects of a system, such as its edge (area network of sensors) are decentralized, system asa whole still involves centralized control in a form of gateway(s) and server(s). Apart fromthe mandatory properties that determine if a system is self-organized, there are severalother characteristics that can appear in it and determine its very important properties[48]:

• Nonlinearity and complex interactions: Self-organizing systems commonly displaycomplex and nonlinear behavior that cannot be understood by separately examiningits components. Such a nonlinear dynamics enables them to better adapt to a largerrange of environmental conditions, and even a small fluctuation can cause a signifi-cantly different final result. Nonlinear systems’ behavior can be achieved by addingup the nonlinear behaviors of the individual components of the system. An exampleof numerous water pipes flowing water into an irrigation bin is a good visualization.Each of them is characterized by a nonlinear behavior, which ultimately results thatthe overall flow of the system also exhibits nonlinear properties.

• Decentralized control: This property has already been discussed in the context ofweak and strong self-organization. To summarize, in weak systems self-organizationis a product of central control, while in the strong ones control is distributed overthe whole system.


• Simple behaviors and local interactions: This property is one of the main reasonswhy self-organized systems are getting so much attention. It means that systemcomponents with simple behaviors, local interactions, and limited perception abilities(components do not have global view of the system, but rather local view with severalof its neighbors) are able to achieve significantly more complex results. For example,molecule in the mentioned Benard example influences only several of its neighboringmolecules it collides with.

• Robustness, resilience: Self-organizing systems consisting of large number of compo-nents can be robust, i.e., immune to errors or perturbations from the environment. Ifone ant is removed from the colony, the result of harvesting food will nevertheless bethe same. Reason for robustness is a redundancy inherent in such systems, becausethe remaining components of the system cover for the ones that failed.

• Emergent properties: The emergent outcome of the system self-organizing processescan be a structure, pattern, behavior, or some other system properties that cannotbe reduced to its basic elements. In the Benard example, resulting cells are createdas a result of a direction of molecules’ rotation. This emergent property cannot beobserved in independent molecules.

6.2. Self-organization in biological systems. The focus of this paper are certainself-organizing principles observed in natural systems, particularly biological systems.Natural systems are dictated by nature in contrast to business and economic systemswhich are governed by business and market laws, and can be broadly divided into physicalsystems, biological systems, and social systems [48]. Research field that investigatesmodels and methods inspired by nature is usually termed as natural computing. Thishighly interdisciplinary field connects biology and computing science, especially in termsof information processing.The study of self-organizing systems has been initiated in 1953 with the work done by

P. P. Grasse [50]. He studied insect societies and found out they achieve order withoutany central point of control. Self-organizing properties have over time been observed inmany natural systems, from insect colonies to flocks of birds and schools of fish, and theyhave inspired solutions to many practical problems. S. Camazine et al. in [51] providemany outstanding examples of such self-organization in biological systems. Foraging ants[52] that explore their environment and seek food can be used to find the shortest possiblepath given the environmental constraints. Swarms find their appliance in fixed and mo-bile networks systems management [53], routing and load balancing [54], or security [55].Self-organization in Peer-to-Peer (P2P) and Mobile ad hoc Network (MANET) includinggossip-based overlay topology management [56] or decentralized techniques for routing,updates, and identity management are proposed in [57]. Network coordination problemshave been solved using techniques inspired by swarm-based models [58] or mimicking in-sect foraging behavior [59]. Self-organizing sensor networks are used in numerous civil andmilitary applications, and recent research in this area focuses on routing, synchronization,and power conservation [60] or decentralized collaborative detection of events [61]. Thereare numerous other areas and applications that have been the subject of self-organizationthemed research, so for more examples consult the associated references section.

6.3. Gossip. Insect social behavior and its self-organizing properties have inspired manyresearch activities in various scientific areas. Some examples were given in the previoussection. Human social interactions are also observed in the context of natural systems theybelong to. People can be part of incredibly complicated social structures, many of whichdemonstrate self-organizing properties. Apart from the vertebrate neural and immures


systems, human social interactions have attracted much interest of the scientific commu-nity and have inspired various practical applications. Social networks have been a subjectof an intensive sociological research for decades, and until only recently have linked upwith mathematical and computational studies. Although initially not expected, researchconfirmed that networks which were a result of social interactions, although largely self-organizing, are not random graphs. Experiments inspired by the popular Erdos number(the number of collaborative links through scientific publications needed to connect tofamous mathematician Paul Erdos) suggested that only six links were needed to connectalmost everyone in the world [62]. Networks that describe human social interactions sharedistinct class of mathematical properties which positions them between random graphsand fully ordered networks. They have become known as the small-world networks andhave inspired first important area of research based on human social interactions [63].

Simple interactions between humans known as gossip have inspired the second impor-tant research area based on human social behavior. Gossip is characterized by its highspeed: information spreads very quickly, analogous to an epidemic infection. The differ-ence in the latter case is that viruses play the role of disseminated information, but theunderlying mechanics is the same. So, it is not uncommon when describing properties ofa gossip protocol to accept and use epidemiologic terminology. More details on humangossip can be found in [64]. The use of epidemic (gossip) algorithms has been explored inapplications such as information dissemination among a large number of nodes [65], failuredetection [66], resource discovery and monitoring [67], data aggregation [68], and databasereplication [69]. The latter scenario represents the first practical application of gossip, andhas introduced several basic variants of gossip-based information dissemination models.

In the context of this paper, our interest lies in studying various information dissemina-tion techniques that can be implemented in computer and communication systems, theirproperties, advantages and disadvantages, and the possibility of using such protocols inthe implementation of a network-centric networking in a decentralized environment.

Information dissemination. Distributed systems today are large-scaled and highlydynamic in nature [70]. The traditional client-server model is not an adequate solution toconnect large number of nodes, enable information exchange between them, and handlefailures or dynamic memberships in the system. The problem of reliability and scalabilitythat occurs has to be tackled with radically different organization: P2P computing model.Each node there can potentially assume the role of either a client or a server. Scalabilityis achieved because the load is balanced between all nodes in the network, which preventsthe occurrence of central point of failure and bottleneck. The problem of implementinginformation exchange on an application level is not straightforward, and in the context ofM2M systems that are currently being standardized represents an active area of research.

Epidemic dissemination protocols are simple, scalable, and easy to deploy. Based onthe same ideas as the spread of disease on a population, these protocols also show similarproperties. Just the way epidemics shows resilience in case of failures (some infect peopleto die before they are able to spread a disease), epidemic protocols are also able toovercome link/node failures. Their scalability is based on the fact that there is not a singlepoint of failure. In case of a node/link failure, other nodes will still be able to spreadinformation through different routes. Each node in a system typically communicates onlywith its neighboring nodes because it lacks the global picture of the system. Complete viewof all nodes in a system is not a realistic assumption in a large-scale network, especiallyin ad hoc networks which feature many joining and departures. Also, maintaining anup-to-date membership view without the irregularities is almost impossible, and bringsunnecessary message overload in the network. However, some earlier implementations ofepidemic protocols featured such an approach [69].


Figure 7. SI gossip model (anti-entropy)

Two basic epidemic protocol variants known in the literature are the SI and SIR model.According to the terminology of epidemiology, each node can be in one of the followingstates [48]:

• Susceptible (S): The node is not yet infected, i.e., it does not know the update;• Infected (I): The node knows the update and is trying to infect others around him,i.e., spread the update to other nodes;

• Removed (R): The node has developed immunity or died, i.e., it is no longer spreadingthe information.

The first, SI model (also called anti-entropy), includes only two possible states. Nodeis either susceptible or infected, and behaves accordingly. Spreading of information toits neighbors is happening according to the prototype showed in Figure 7. Node spreadsinformation once in each ∆ time units. This period is called a gossip cycle. It choosesa random node p out of the set of all available nodes. Next important choice is theselection of push or pull parameter. At least one of them has to be true, so the availablecombinations are push, pull, or push-pull gossip. In push gossip, susceptible nodes arepassive, while in the two remaining cases each node is active. All the three variantseventually infect the entire network, but offer different performance while doing so. Thepush-pull model works faster than the other two variants.A bit more complex model, the SIR epidemics model, was developed to cope with some

of the problems encountered by the SI. Anti-entropy ignores one very important aspectof such systems: termination. SI push protocols never terminate, while pull have theability to do so, but need to know the list of all available updates in advance which israrely known in practice. Therefore, a more advanced epidemic model that solves thisproblem was developed. Algorithm showed on Figure 8. is very similar to the one ofSI model. Major difference is the onFeedback procedure that allows a transition to the


Figure 8. SIR gossip model (rumor mongering)

removed state with the probability of 1/k, where k stands for average number of times anode sends update to his peer that already has it. This mechanism, also called “rumormongering”, is based on “hot rumors”. It means when a node receives a new update itis considered “hot”, so it tries to send it to all nodes in his peer list. However, whenhis peer informs it that it already possesses this update, the sender switches its state toR with the mentioned probability. This behavior causes a rather important implication:depending on how fast the system converges to inactive state (all nodes are R), there is anexplicit probability that the complete dissemination of information will not be achieved.In other words, rumor mongering does not assure that the sent updates will be receivedby all its intended destinations. For more details on SI and SIR epidemic models consultaccompanied literature [48,69].

Based on the brief analysis, it is safely to assume that an actual distributed systeminformation is spread using rumor mongering, but the occasional run of anti-entropyis necessary to take care of possible undelivered updates. In chapter 5, we discussedtwo M2M scenarios and analyzed their compatibility with the network-centric approach.The second one involves a possible farm field implementation and features decentralizednetwork of M2M devices with sensors scattered on a possibly large area. The use of low-coverage technology such as ZigBee encourages the usage of self-organizing mechanisms for


information dissemination. Source node highlighted in the figure cannot send informationto the highlighted destination directly as it does not possess the necessary technology todo so. It has an ability to send information to its neighboring nodes, and then rely onthe fact that they will do the same. Obviously, self-organizing methods such as gossip(epidemic) protocols are almost a natural fit for these occasions. They offer a robust andscalable way of propagating information in distributed systems, and based on the briefoverview of their basic features, are a compatible solution for implementing “informationgrid” of desired properties (universal access to information, information bus that connectsall M2M devices, ability to continuously adapt to changes (dynamic network topologies,membership changes, etc.)) as proposed by the network-centric operations.Applying gossip information dissemination in distributed systems. Epidemic

algorithms have been studied theoretically because they are based on sound mathematicalfoundations [70]. There are several non-trivial issues that have to be carefully analyzed(e.g., analysis in [71] discovers that gossip’s robustness crucially depends on a handful ofassumptions that are often left unspoken) and resolved before epidemic algorithms canbe applied in a practical distributed environment [72].Every node in a network is a potential relay in a dissemination process and has a buffer

of capacity b, sends messages limited number of times t to a randomly selected set ofnodes of size f (fanout). Probabilistic guarantees of message delivery are directly relatedto the value of dissemination parameters, and epidemic algorithms in this sense displaybimodal behavior: there is a clear threshold in the value of these parameters and thehigh probability of a reliable delivery. The reliability of message delivery is achieved withredundancy and randomization in order to bypass potential node and link failures. Whenimplementing epidemic information dissemination in a practical setting, there are severalimportant design constraints that need to be taken into account, and several importantassumptions about the environment in which the protocols operate:

• Membership: The goal of the membership issue is to define how a node choosesnodes it knows, and then sends them information that is being disseminated. Thishas impact on the performance of the information dissemination process. The origi-nal ideas discussed in [69] assumed that every node in a network knows every othernode. There are two important constraints for such an assumption. Firstly, mem-bership information grows as the network grows. Secondly, maintaining consistentmembership information about every node in a network imposes an extra networkload, especially in dynamic environments (e.g., P2P networks). Therefore, becauseof the scalability requirements, each network node in a network in a practical exam-ple typically has only partial view of the network. The tradeoff between small andlarge views is also a tradeoff between scalability and reliability. Small views scalebetter, while large views reduce the chance that node becomes isolated.

• Network awareness: So far, discussions regarding epidemic algorithms assumed thatall nodes are equally reachable, i.e., did not take into account the underlying networktopology. In practice, such ignorance can be very costly: it is not an unlikely scenariowhere information is disseminated from one node to a very close node via a thirdremote node. Most solutions proposed to address this issue are based on hierarchicalorganizations which mimic the network topology [73] or use an administration systemaware of the actual hierarchy [67].

• Buffer management: Simple epidemic protocol is based on the following sequenceoff steps: a node receives an update message and stores it in a buffer of capacity b,forwards that message a limited number of times t, each time to a randomly selectedset of nodes of size f . Buffer’s role is to ensure that every message is stored long


enough to enable it being sent a sufficient number of times to achieve an acceptablereliability of information dissemination. Strategies on what to do when the bufferis full: nodes in a conservative strategy simply reject and drop new messages, whilea dynamic strategy drops old messages (those that have been forwarded a sufficientnumber of times) according to certain criteria. The goal of these strategies is toensure memory usage optimization and/or resource scalability, while maintaining anacceptable reliability. An example is a strategy which classifies messages accordingto their age, i.e., the number of times a message has been sent to another node.When a buffer is full, instead of dropping a new message, node chooses to dropthe oldest one. Similar result can be achieved using application semantics [74] thatdefines an obsolescence relation: message m1 makes message m2 obsolete in a sensea node that receives m1 does not need m2 anymore. This approach can be combinedwith the age-based priority.

• Message filtering: So far, all discussions assumed that every node in a networkis equally interested to receive all messages. However, nodes could be partitionedinto distinct groups according to their interest, and messages disseminated accord-ingly. Epidemic dissemination scheme can be enhanced with filtering capabilitiesthat switch randomized neighbor selection scheme with a heuristic to privilege onlyinterested nodes. The goal is to increase probability that a node receives a messageonly it is interested in. Non-randomized solutions take into account node’s interestsand dynamically evaluate exchanged messages based on their contents. The designof such a filtering mechanism is not a trivial problem. Providing knowledge on node’sinterest in a decentralized manner is a first major issue. Secondly, even when a nodeknows that a certain message is not interesting to its neighboring node, there is stilla genuine possibility that this node might be critical in reaching message’s intendeddestination. Obviously, making nodes communicate only with nodes of the sameinterest is hard, because nodes only know subsets of the other nodes in the network.In addition, achieving network awareness together with message filtering is even amore complex problem. Approach presented in [75] arranges nodes in a form of ahierarchy according to their geographical distances, while their interests are groupedat each level of hierarchy at the same time, and achieves very good performance.

7. Conclusion. M2M systems due to their diversity still faced with management andcoordination challenges. NCO as an information management concept is a way of max-imizing the value of M2M solutions, moving from centralized to decentralized controland reaching the maximum utilization of available information. It is shown that M2Menvironment is capable of implementing key features of such an approach. Apart fromthe described M2M e-Health system where a network-centric networking, although not acompletely natural fit, offers enrichment of the current system’s operations, there are alsoM2M systems with inherently decentralized organizations that are so far mostly ignored bythe standardization efforts and would benefit even more from the inclusion of NCO ideas.Implementing such an approach proposes the development of an interconnected, robustand dynamic information grid within the M2M system network infrastructure that willincrease shared situational awareness and allow self-synchronization of connected smartdevices. As it was discussed in the paper, natural self-organizing systems offer a viablesolution: gossip (epidemic) protocols. They have already been implemented in variousdistributed systems and display many desired properties. However, there are several non-trivial issues that have to be carefully analyzed before initiating their customization forspecific practical purposes.


Future work, according to the presented material, offers many possibilities. Furtherresearch on the topic of information dissemination, customization of gossip protocols thatwould be able to take advantage of the specific properties of M2M applications, or thedevelopment of semantic support for information exchange and social interactions betweenM2M machines are just few of the many available options.

Acknowledgment. This work was supported by two research projects: “Content De-livery and Mobility of Users and Services in New Generation Networks” (036-0362027-1639), funded by the Ministry of Science, Education and Sports of the Republic of Croatiaand “Looking to the Future”, funded by Croatian Post and Electronic CommunicationsAgency. The authors also gratefully acknowledge the helpful comments and suggestionsof the reviewers, which have improved the presentation.

REFERENCES

[1] Smart Devices and Services Connected by CDMA2000, Harbor Research, Inc., 2010.[2] M2M: Growth Opportunities for MNOs in Developed Markets (Sample Pages), Mobile Market De-

velopment Ltd., 2010.[3] M. Alendal, Operators need an ecosystem to support 50 billion connections, Ericsson Review, no.3,

2010.[4] A. Cox, M2M: Connecting the devices, M2M & embedded strategies, telematics, CE, mHealth,

metering & smart buildings 2011-2016, Juniper Research, 2011.[5] ETSI Technical Specification 102 689: M2M Service Requirements, v2.0.3, ETSI, 2012.[6] 3GPP Technical Report 22.868: Study on Facilitating M2M Communication in 3GPP Systems,

v8.0.0, 3GPP, 2007.[7] N. Tekbiyik and E. Uysal-Biyikoglu, Energy efficient wireless unicast routing alternatives for

machine-to-machine networks, Journal of Network and Computer Applications, vol.34, no.5, pp.1587-1614, 2011.

[8] G. Lawton, Machine-to-machine technology gears up for growth, IEEE Computer, vol.37, no.9,pp.12-15, 2004.

[9] D. Katusic and G. Jezic, Network-centric operations in machine-to-machine networks, Lecture Notesin Computer Science, vol.7327, pp.474-483, 2012.

[10] F. Dressler, Network-centric actuation control in sensor/actuator networks based on bio-inspiredtechnologies, Proc. of the 2006 IEEE International Conference on Mobile Ad Hoc and Sensor Sys-tems, pp.680-684, 2006.

[11] S.-Y. Lien, K.-W. Chen and Y. Lin, Towards ubiquitous massive accesses in 3GPP machine-to-machine communications, IEEE Communications Magazine, vol.49, no.4, pp.66-74, 2011.

[12] K. Chang, A. Soong, M. Tseng and Z. Xiang, Global wireless machine-to-machine standardization,IEEE Internet Computing, vol.15, no.2, pp.64-69, 2011.

[13] oneM2M, http://www.onem2m.org/, 2012.[14] ETSI, http://www.etsi.org/website/homepage.aspx, 2012.[15] ETSI M2M Technical Committee, http://www.etsi.org/Website/Technologies/M2M.aspx, 2012.[16] ETSI Technical Report 102 691: Smart Metering Use Cases, v1.1.1, ETSI, 2010.[17] ETSI Technical Report 102 732: Use Cases of M2M Applications for eHealth, v0.4.1, ETSI, 2011.[18] ETSI Technical Report 102 857: Use Cases of M2M Applications for Connected Consumer, v0.3.0,

ETSI, 2010.[19] ETSI Technical Report 102 898: Use Cases of Automotive Applications in M2M Capable Networks,

v0.4.0, ETSI, 2010.[20] ETSI Technical Report 102 897: Use Cases of M2M Applications for City Automation, v0.1.1, ETSI,

2010.[21] ETSI Technical Report 102 725: M2M Definitions, v0.8.0, ETSI, 2012.[22] ETSI Technical Specification 102 690: M2M Functional Architecture, v2.0.9, ETSI, 2012.[23] ETSI Technical Report 101 603: 3GPP Interworking, v0.0.5, ETSI, 2012.[24] ETSI Technical Report 103 107: 3GPP2 Interworking, ETSI, 2012.[25] ETSI Technical Specification 103 092: OMA DM Compatible Management Objects for ETSI M2M,

v1.2.1, ETSI, 2013.[26] ETSI Technical Specification 103 903: BBF TR-069 Compatible Data Model, v2.0.3, ETSI, 2012.


[27] ETSI Technical Report 102 966: Interworking between the M2M Architecture and M2M Area NetworkTechnologies, v0.2.0, ETSI, 2012.

[28] 3GPP, http://www.3gpp.org/About-3GPP, 2012.[29] 3GPP Technical Specification 22.368: Service Requirements for MTC, v11.3.0, 3GPP, 2011.[30] 3GPP Technical Report 22.988: Study on Alternatives to E.164 for MTC, v2.0.0, 3GPP, 2012.[31] 3GPP Technical Specification 02.30: Man-machine Interface of the Mobile Station, v7.1.1, 3GPP,

2002.[32] 3GPP Technical Specification 22.030: Man-machine Interface of the User Equipment, v12.0.0,

3GPP, 2012.[33] 3GPP Technical Report 23.888: System Improvements for MTC Communication, v1.6.0, 3GPP,

2011.[34] 3GPP Technical Report 33.812: Feasibility Study on the Security Aspects of Remote Provisioning

and Change of Subscription for M2M Equipment, v9.2.0, 3GPP, 2010.[35] 3GPP Technical Report 33.868: Security Aspects of Machine-Type Communications, v0.10.0, 3GPP,

2012.[36] 3GPP Technical Report 37.868: RAN Improvements for Machine-Type Communications, v11.0.0,

3GPP, 2011.[37] 3GPP Technical Report 43.868: GERAN Improvements for Machine-Type Communications, v12.0.0,

3GPP, 2012.[38] A. K. Cebrowski and J. J. Garstka, Network-centric warfare: Its origin and future, Proceedings

Magazine, pp.28-35, 1998.[39] NIS23, Net-Centric Information & Integration Services for Security System, 2009.[40] D. S. Alberts, J. J. Garstka and F. P. Stein, Network Centric Warfare: Developing and Leveraging

Information Superiority, 2nd Edition, CCRP, 2001.[41] Network Centric Warfare, Department of Defense Report to Congress, 2001.[42] ETSI Technical Report 101 584: Study on Semantic support for M2M Data, v0.5.0, ETSI, 2012.[43] N. J. Wesensten, G. Belenky and T. J. Balkin, Cognitive readiness in network-centric operations,

Parameters, vol.35, no.1, pp.94-105, 2005.[44] J. S. Bay, Disruptive effects of net-centricity on command and control, Proc. of the 13th ICCRTS

“C2 for Complex Endeavors”, 2008.[45] J. L. Groh, Network-centric warfare: Leveraging the power of information, USAWC Guide to Na-

tional Security Issues, vol.1, pp.323-338, 2008.[46] D. von Lubitz and N. Wickramasinghe, Healthcare network centric operations: The confluence

of E-Health and E-Government, Global E-Government: Theory, Applications and Benchmarking,pp.127-147, 2007.

[47] D. C. Schmidt, A. Corsaro and H. van’t Hag, Addressing the challenges of tactical informationmanagement in net-centric systems with DDS, PrismTech Corporation, 2008.

[48] G. D. M. Serugendo, M.-P. Gleizes and A. Karageorgos, Self-Organising Software: From Natural toArtificial Adaptation, Springer-Verlag, Berlin, 2011.

[49] F. Heylighen, The science of self-organization and adaptivity, Knowledge Management, Organiza-tional Intelligence and Learning, and Complexity, pp.253-280, 2001.

[50] P. P. Grasse, La reconstruction du nid et les coordinations interindividuelles chezbellicositermesnatalensis ectubitermes sp la theorie de la stigmergie: Essai d’interpretation du comportement destermites constructeurs, Insectes Sociaux, vol.6, no.1, pp.41-80, 1959.

[51] S. Camazine, J. L. Deneubourg, N. R. Franks, J. Sneyd, G. Theraulaz and E. Bonabeau, Self-Organization in Biological Systems, 2nd Edition, Princeton University Press, 2003.

[52] J. L. Deneubourg, S. Goss, N. Franks, A. Sendova-Franks, C. Detrain and L. Chretien, The dynamicsof collective sorting robot-like ants and ant-like robots – Simulation of animal behaviour, Proc. ofthe 1st International Conference of Simulation of Adaptive Behaviour, pp.356-363, 1991.

[53] S. Brueckner and H. V. D. Parunak, Self-organising MANET management, Lecture Notes in ArtificialIntelligence, vol.2977, pp.20-35, 2004.

[54] A. Montresor, H. Meling and O. Babaoglu, Messor: Load-balancing through a swarm of autonomousagents, Lecture Notes in Artificial Intelligence, vol.2530, pp.125-137, 2003.

[55] N. Foukia, IDReAM: Intrusion detection and response executed with agent mobility, Proc. of Inter-national Conference on Autonomous Agents and Multi-Agent Systems, New York, USA, pp.264-270,2005.

[56] M. Jelasity and O. Babaoglu, T-man: Gossip-based overlay topology management, Lecture Notes inComputer Science, vol.3910, pp.1-15, 2006.


[57] K. Aberer, A. Datta and M. Hauswirth, P-grid: Dynamics of self-organizing processes in structuredpeer-to-peer systems, Lecture Notes in Computer Science, vol.3485, pp.137-153, 2005.

[58] G. D. Caro and M. Dorigo, Ant colonies for adaptive routing in packet switched communication net-works, Proc. of the 5th International Conference on Parallel Problem Solving from Nature, London,UK, pp.673-682, 1998.

[59] R. Schoonderwoerd, O. Holland and J. Bruten, Ant-like agents for load balancing in telecommuni-cations networks, Proc. of the 1st International Conference on Autonomous Agents, Los Alamitos,USA, pp.209-216, 1997.

[60] K. Mills, A brief survey of self-organization in wireless sensor networks, Wireless Communicationsand Mobile Computing, vol.7, pp.823-834, 2007.

[61] J. L. Fernandez-Marquez, J. L. Arcos and G. D. M. Serugendo, A decentralized approach for detectingdynamically changing diffuse event sources in WSN environments, Proc. of the 4th InternationalConference on Self-Adaptive and Self-Organising Systems, Los Alamitos, USA, 2010.

[62] D. Watts, Six Degrees: The Science of a Connected Age, Vintage, London, 2004.[63] D. Watts, Small Worlds: The Dynamics of Networks Between Order and Randomness, Princeton

University Press, 1999.[64] A. J. Kimmel, Rumors and rumor control: A manager’s guide to understanding and combating

rumors, Taylor & Francis Group, 2004.[65] R. M. Karp, C. Schindelhauer, S. Shenker and B. Vocking, Randomized rumor spreading, Proc. of

the 41st Annual Symposium on Foundations of Computer Science, Berkeley, USA, pp.565-574, 2000.[66] R. van Renesse, Y. Minsky and M. Hayden, A gossip-style failure detection service, Proc. of the

IFIP International Conference on Distributed Systems Platforms and Open Distributed Processing,London, UK, pp.55-70, 1998.

[67] R. van Renesse, K. P. Birman and W. Vogels, Astrolabe: A robust and scalable technology fordistributed system monitoring, management, and data mining, ACM Transactions on ComputerSystems, vol.21, no.2, pp.164-206, 2003.

[68] M. Jelasity, A. Montresor and O. Babaoglu, Gossip-based aggregation in large dynamic networks,ACM Transactions on Computer Systems, vol.23, no.3, pp.219-252, 2005.

[69] A. Demers, D. Greene, C. Hauser, W. Irish, J. Larson, S. Shenker, H. Sturgis, D. Swinehart andD. Terry, Epidemic algorithms for replicated database management, Proc. of the 6th Annual ACMSymposium on Principles of Distributed Computing, Vancouver, Canada, pp.1-12, 1987.

[70] N. T. J. Bailey, The Mathematical Theory of Infectious Diseases and Its Applications, 2nd Edition,Griffin, London, 1975.

[71] L. Alvisi, J. Doumen, R. Guerraoui, B. Koldehofe, H. Li, R. von Renesse and G. Tredan, How robustare gossip-based communication protocols? ACM SIGOPS Operating Systems Review – Gossip-Based Computer Networking Archive, vol.41, no.5, pp.14-18, 2007.

[72] P. T. Eugster, R. Guerraoui, A.-M. Kermarrec and L. Massoulie, From epidemics to distributedcomputing, IEEE Computer, vol.37, no.5, pp.60-67, 2004.

[73] A.-M. Kermarrec, L. Massoulie and A. J. Ganesh, Probabilistic reliable dissemination in large-scalesystems, IEEE Transactions on Parallel and Distributed Systems, vol.14, no.3, pp.248-258, 2003.

[74] J. Pereira, L. Rodrigues and R. Oliveira, Semantically reliable multicast algorithms, IEEE Transac-tions on Computers, vol.52, no.2, pp.150-165, 2003.

[75] P. T. Eugster and R. Guerraoui, Probabilistic multicast, Proc. of the 2002 International Conferenceon Dependable Systems and Networks, pp.313-324, 2002.

implementing self-organizing information … · and self-synchronization of autonomous and...

Documents